Was this newsletter forwarded to you? Sign up to get it in your inbox.
Fifteen years ago, I took my first online course on edX, a learning provider founded by MIT and Harvard. At the time, edX was cutting edge and I binge-watched a ton of courses. (I still love and recommend Michael Sandel’s course Justice, but that’s beside the point.)
I found myself recalling that time when I took four courses earlier this summer to earn our team Anthropic certification. After working closely with the Anthropic team for more than a year as alpha testers of unreleased models, we wanted to take our partnership farther. I went into the certification process curious about what level of instruction they’d provide and how they’d define their own tools. What I came away with was new insight into the state of AI development and industry standards right now.
AI development is still moving at a breakneck pace, but new releases no longer upend our work the way they used to. Anthropic’s Opus 4.5 and OpenAI’s GPT-5.5 marked a turning point as the models could reliably perform the tasks we provide, giving each subsequent release a dependable base to build on. The fact that Anthropic can even release a training course for its foundational tools like Claude Code and agent skills is a milestone for the AI era. (This time last year, I was constantly reworking the curriculum for our training sessions every three to six weeks when a new, more capable model came out.)
More stable models have allowed people to build AI workflows around their individual roles and careers. Anthropic’s training takes the opposite tack—it teaches foundational concepts. But the greatest value of these courses may not be sparking new ideas for the tools we use day to day. Instead, they offer an opportunity for the industry to align around terms and definitions—a heavy lift on its own.
Shared reference points, not playbooks
To become Anthropic certified, companies are required to take four courses: Introduction to Agent Skills, Building with Claude API, Introduction to Model Context Protocol (MCP), and Claude Code in Action. The full course load takes roughly 10 to 15 hours per person.
Each online training involves reading through text-based how-tos interspersed with short videos and sample exercises, much like the many online courses I’ve taken previously—including in those edX days.
None of the four courses delivers a playbook for transforming workflows with AI—mapping workflows, identifying tasks ripe to streamline with AI, and codifying tasks into skills. Instead, they offer common definitions: This is what Anthropic says a skill is; this is what an MCP is; this is how an API works.
Definitions may seem like a starting bid for a training, but they are critical when you’re creating the foundation for tools that you want hundreds of millions of people to use, which Anthropic is—this certification is literally AI 101.
If Anthropic’s definition of an MCP diverges from an individual company’s, that difference could hurt effective AI implementation. Having a shared language is important when we’re still figuring out AI adoption as an industry.
I felt the benefit myself. I’m not often building MCPs, so despite how often we use them at Every, my understanding only clicked into place when the course walked through the endpoints and logic underneath them.
The training also gave our team a shared reference point. Getting roughly a third of the company to take this course catalyzed a conversation internally about how we learn. While the training is good, the Every team consensus is that Anthropic’s documentation, a publicly available collection of guides and resources for common use cases, is much better. If you really want to understand how Claude works, skip the videos—the documentation is the gold standard.
One-size-fits-all training for a rapidly specializing field
While the pace of change at the model level has slowed, the user experience of the software around AI models is changing faster than ever. Unfortunately, the courses reflected that.
The Building with Claude API course uses a Sonnet model that’s no longer available in the API, and the MCP course doesn’t mention Anthropic’s own MCP builder skill. I don’t mean that as a criticism of Anthropic. It’s a symptom of how fast things are changing, and an occupational hazard when trying to capture a still-developing process for educational purposes.
The courses also don’t take your role or experience level into account, or assess what you already know before moving into the material. Whether you’re a CFO, an engineer, or an intern, your experience working through certification is identical. This feels like a missed opportunity when AI is already very good at personalizing content to a viewer. But the bigger issue is that AI adoption has become increasingly specialized: The tools and workflows I use as an AI implementation leader look nothing like those of an engineer and or a hedge fund manager. The courses don’t account for that shift—none included examples grounded in actual workflows. A training built for hedge fund managers, for instance, could show a skill applied to a financial model. Instead, Anthropic has built a training program general enough to serve everyone—and that might explain the split in experiences across the 10 people on our team who went through the courses.
Some found the selection of courses puzzling. Mike Taylor, head of evals at Every, thought the Building with Claude API course was long, technical, and, in his view, only relevant to developers. That sits oddly with the fact that most of the opportunity for AI tooling education is with non-technical people. “MCPs were already touched on in the API course, and you don’t need to know much about that unless you’re building one yourself. Even the developers I know aren’t building their own MCPs,” Mike says.
A few like Becky Isjwara, head of social media at Every, found the certification training in general largely unnecessary. She has previously built and shipped usable web apps through vibe coding only and didn’t see how more specific technical information would change her results. “I think it just proves that anybody can code if they want to get their hands dirty and start building stuff,” she says. “You don’t need technical knowledge to ship anything.”
By contrast, Yash Poojary, a growth engineer at Every, finished the API course and immediately proposed we build it into onboarding because it was so helpful. “There’s useful information for anyone going deeper or meeting the material for the first time,” says Lee Knowlton, an engineer. Already knowledgeable about the technical terms, he benefited from peeking under the hood at how Anthropic thinks about system design and performance. Plus, Lee found the section on prompt evaluation systems more compactly explained than anything he’d seen elsewhere. I myself learned a lot about MCPs.
A valuable-enough training with more value still to come
The courses haven’t fundamentally changed how we run the consulting practice, nor is it something I’d recommend to the time-poor executives we work with. But that’s less of a judgment on the quality of the courses and more of a description of where the technology is at. And most of the Every team still said they were glad to have taken the required courses.
We now have a shared and in-depth understanding of how the labs define the key tools we build with every day. The training gave everyone a common vocabulary around AI’s core tools—and made the whole thing feel less intimidating. That’s exactly what the courses excel at and what makes AI education worthwhile even for a fast-moving industry.
By that measure the Anthropic certification process succeeds. It just doesn’t go further than that, and right now, maybe no one can.
Natalia Quintero is the head of consulting at Every. You can follow her on X at @NataliaZarina and on LinkedIn. To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.
Thanks to Tom Matsuda for editorial support.
Everyone’s a builder now. Every All Access gets you the full membership plus the Builder Pack—$9,000+ in credits for the tools we build with.
